全景视觉中基于轮廓特征的运动目标识别方法  

A Moving Object Recognition Method Based on Contour Features in Panoramic Vision

作　　者：查筱菲 单单 ZHA Xiao-fei;SHAN Dan(Tongji University,Shanghai 201200,China)

机构地区：[1]同济大学,上海201200

出　　处：《计算机仿真》2025年第3期516-520,共5页Computer Simulation

摘　　要：由于运动目标可能会被其它物体或背景遮挡,导致目标在图像中不完整或部分不可见,加大了目标识别难度。为了提高目标识别率与识别效果,提出一种全景视觉中基于轮廓特征的运动目标识别方法。利用高斯背景模型与五帧差分法结合的方式实现视频图像的差分处理,根据差分处理结果提取目标所具备的边界矩特征及形态学特征。通过相似度计算对图像中的单一运动目标展开识别跟踪,同时构建SVM分类器实现运动目标的整体识别。经实验验证可知,利用所提方法开展运动目标识别时受遮挡因素的影响较小,具有较高的识别率,且识别效果较好。At present,moving targets may be obscured by some objects or backgrounds,causing the target to be incomplete or partially invisible in images.In order to improve the target identification rate and effect,this paper proposed a method of moving target identification in panoramic vision based on contour features.Firstly,the Gaussian background model was combined with the five-frame difference method to perform the differential analysis in video images.Then,based on the differential processing result,boundary moment features and morphological features possessed by the target were extracted.Through similarity calculation,single moving targets in the image were identified and tracked.At the same time,an SVM classifier was constructed to achieve the overall identification of moving targets.Experiment verification shows that during the moving target identification,the proposed method is less affected by occlusion factors,and has a high identification rate as well as a good identification effect.

关 键 词：全景视觉 轮廓特征 特征提取 目标识别 背景分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]